GLSL loop wrong behaviour - opengl

Question

We have the following piece of GLSL code:
#version 120
#define LIGHT_COUNT 2
...
varying vec3 v_light_tan[LIGHT_COUNT];
varying float v_attenuation[LIGHT_COUNT];
uniform vec3 light_position[LIGHT_COUNT];
void main(void) {
...
for ( int index = 0; index < LIGHT_COUNT; index++ )
{
float distance = length( light_position[index] - world_pos );
v_attenuation[index] = 1.0 / ( 1.0 + 0.05 * distance * distance);
vec3 light_loc = vec3(nv_m_model_inv * vec4 (light_position[index], 1.0) );
vec3 light_dir_loc = normalize( nv_position - light_loc );
v_light_tan[index] = vec3(
dot( nv_tangent.xyz, light_dir_loc ),
dot( bitangent, light_dir_loc ),
dot( nv_normal, light_dir_loc )
);
}
I've read a lot about problems with loops, so I went with a hardcoded loop. But even this simple code seems to fail. The above works for LIGHT_COUNT = 1, but for 2, the value of v_attenuation[0] gets corrupted. Surprisingly, changing the code to:
for ( int index = 0; index < LIGHT_COUNT; index++ )
{
float distance = length( light_position[index] - world_pos );
v_attenuation[index] = 1.0 / ( 1.0 + 0.05 * distance * distance);
}
for ( int index = 0; index < LIGHT_COUNT; index++ )
{
vec3 light_loc = vec3(nv_m_model_inv * vec4 (light_position[index], 1.0) );
vec3 light_dir_loc = normalize( nv_position - light_loc );
v_light_tan[index] = vec3(
dot( nv_tangent.xyz, light_dir_loc ),
dot( bitangent, light_dir_loc ),
dot( nv_normal, light_dir_loc )
);
}
... ergo two separate loops, works. Adding #pragma optionNV(unroll all) works as well.
Both of those seem hacky though, for something that is such a simple loop example. Is GLSL that broken, or did I miss some rule or guideline of writing GLSL loops?
If relevant:
OpenGL Vendor : NVIDIA Corporation
OpenGL Renderer : GeForce GTX 690/PCIe/SSE2
OpenGL Version : 4.4.0
OpenGL GLSL Version : 4.40 NVIDIA via Cg compiler